grid2viz icon indicating copy to clipboard operation
grid2viz copied to clipboard

Reward graph is always the same for 2 agents while score are different

Open marota opened this issue 4 years ago • 3 comments

In Agent Overview, the reward graph between 2 agent always look identical (except when a time of game over is mentioned). It is weird as it should differ at least sometimes. For instance in the attached agents, the two have different scores overall on scenario jun14_2

image

image

Is the reward curve for each agent properly displayed ?

This might be a resolution problem has we cannot really zoom in very accurately. See proposal of ploting instant & cumulated reward on 2 different plots https://github.com/mjothy/grid2viz/issues/50#issuecomment-712273854

marota avatar Oct 19 '20 15:10 marota

We should indeed see a slight diferrence in the cumulated sum by the end

image

marota avatar Oct 19 '20 16:10 marota

Data to reproduce NeurIPS_1.2.2_test_reward_.zip

Notebook to inspect reward differences Testing_reward_difference.ipynb.zip

marota avatar Oct 19 '20 16:10 marota

This is fixed I think (I did not checked the data in the notebook though) image

vinault avatar Oct 21 '20 22:10 vinault